Research On Friend Recommendation Algorithm Based On Multi-information Source Graph Embedding

From the era of information scarcity to the era of information overload,users face the amount of information that exploded,and it has become more and more difficult for users to select the required information.The emergence of personalized recommendation system has effectively solved the problem.It analyzes the user's historical behavior information,and provides the accurate information for users.As social networks continue to evolve,more and more users like to make new friends in the social network,so friendship becomes an important part of social networks.Many social networking sites also set up a friend recommendation module,the module can find other users who are similar to the target user,then it can recommend new friends for the target user.Collaborative filtering is the most widely used method in the friend recommendation algorithm.The traditional user-based collaborative filtering algorithm analyzes the similarity between users through the user-user relationship matrix,and recommends new friends according to the similarity.However,in real life,the user-user interaction matrix contains a large number of missing values,and when new users join in the social network,we cannot obtain interactive information,so it can cause problems such as sparse data and cold start of new users.Therefore,in order to solve the above problem,the attribute information of the user is integrated as auxiliary information to the algorithm.In this thesis,we propose a friend recommendation algorithm based on multiinformation source graph embedding.First,the algorithm considers both the structural information and attribute information of the node,and excavates the preference of the node from the attribute information.Second,the attribute information and structure information are mapped into low-dimensional vectors through graph embedding methods.The structural proximity and attribute proximity of the network are preserved.The structural proximity captures the global network structure,and the attribute proximity explains the homophily effect.Finally,the two parts of information are aggregated into the multi-layer perceptron for training and the model predicts the probability of a link between nodes.To verify the effectiveness of the proposed algorithm,we performed experiments on three real datasets.The experimental results show that the model does improve the recommendation performance.